Hacker News Reader

Comment by CoolGuySteve | Hacker News Reader

By the time the fab comes online in 2024 (which seems optimistic), 4nm should be a half-step generation or so behind.

TSMC also has a dozen or so fabs in Asia, so it's not clear how the single plant in Arizona is going to meet capacity requirements or how often that plant will be retrofitted with newer equipment.

This plant and the federal subsidies backing it seem more like a way for defense contractors to domestically source relatively recent fab processes over the next couple decades rather than something intended solely for consumer products.

The fact that consumer facing companies are interested in using the fab when it comes online doesn't necessarily mean they'll still be using it 10 years from now unless TSMC keeps it up to date.

View on HN

TSMC expects to begin mass production of 3nm chips in its Taiwan fabs by 2023 Q4: https://seekingalpha.com/article/4546779-tsmc-no-3nm-soon

This isn't my field, but I'm not sure what's the excitement about an Arizona plant that will optimistically start production a year later with an older process.

There is still a lot of demand for the not-latest-generation, "long tail" fab processes. Much of the long tail manufacturing was (is?) what the chip shortage was about, rather than the current generation.

A good example is the automotive industry. It doesn't typically move quickly or frequently onto newer generation processes as that requires R&D time and other expenses. However, these have been some of the worst shortages, with auto manufacturers still impacted by lead times (although, this is starting to clearing up fwiu).

Targeting the current-best process, rather than the next generation, alleviates some compounding of risk that would be incurred by logistics concerns of turning up a new site and putting that new site on a process that they don't yet have full confidence in because it's not seen production yet.

Does "one generation before current" count as "long tail", though?

The chip shortage in products like cars has been for much simpler chips, AFAIK.

The article says Apple, AMD, and Nvidia are looking to source from the new plant. Aren't they canonical examples of companies that are always looking for the latest and greatest? Why would they be interested in an prev-gen chip?

Automotive industry is still using 90nm chips. It's not even the generation before the current one. They have long production runs and require stability, and they also prefer to use standardized parts across many production runs.

Defense also uses older chips. Don't ask what kind of chips are in Amraam missiles or the F-35 -- although the F-35 is getting a technology refresh now (after a 14 year production run).

Only in bleeding edge consumer devices does it make sense to keep changing chips. In other systems, production runs can last 20 years and the life of the asset can be 40 years or more, and you want the same spare parts available throughout the entire expected life of all assets produced. And then when you look at fixed assets, such as thermal power stations, then you are looking at even longer time horizons.

That's what I thought. If consumer products tend to use the bleeding edge, which will be the state-of-the-art chips produced in Taiwan, then I still don't get:

1. What's so exciting about a prev-gen fab possibly being completed in 2024?

2. Who will buy the chips produced by the 4nm Arizona fab, and why?

The only plausible answer offered for 1 in this thread has been that it's too difficult to jump straight into the bleeding edge, and this is the most the US can do to lay the foundation for an eventual catch-up with the state of the art in chip fabrication.

If mission critical applications (military, automotive, aerospace) are using older chips, and smartphones/computers are using the latest, then everything else is using chips between those two extremes... computer peripherals, office machines, toys, audio/video systems, communication equipment, lights, HVAC, anything rechargeable. The list is massive, you name an industry and they will probably be buying 4nm over the next couple decades.

Because overall the us chip production capacity and technical reach has been reducing over time. We need to establish and pull over a much larger amount of production so our strategic needs can be met inside the us. This helps us avoid problems like our key sources being cut off or destroyed with our allies like Taiwan and South Korea, or China does more and more production and technical leadership and decides to squeeze us.

Intel has sinking in at least technical leadership for a long time. It's obviously a priority a need for the us to on-shore more chip prod.

Lower end consumer electronics are fine with chips that aren't the absolute latest. The point isn't that they won't find buyers, but rather that they will earn less money building a fab in a location with much higher labor costs and producing chips that are a bit behind.

My guess is that the decision to build fabs in the U.S. was the result of geopolitical strong-arming by the U.S., but TSMC will still find buyers and that the fab will at least pay for itself.

Starting in 2022, the high-volume iPhone has been switched to the N-1 generation SoC, while the bleeding edge SoC is reserved for the lower volume iPhone Pro/Max.

Many companies (including NVidia and AMD) are adopting chiplet designs for which different parts of the chip can be manufactured on different process nodes. The most performance critical aspects of chips will be manufactured on 3 nm nodes, but most of the chip will be manufactured on 4-10 nm nodes. It's too expensive (in cost/wafer) to use 3 nm for everything.

It’s not just cost: a lot of elements have effectively hit serious diminishing returns below 7nm.

> The article says Apple, AMD, and Nvidia are looking to source from the new plant. Aren't they canonical examples of companies that are always looking for the latest and greatest?

CPUs and GPUs get all the press, but there are dozens of other chips used in even the latest models that use older technology. For all we know Apple could be buying those chips for USB cables, or pencils.

The PS5 is based on a 7nm architecure.

You don't start at the cutting edge immediately. You start where it is better understood. Then you can still pay the enormous investments needed for getting to cutting edge and furthering it, if you want. Also, from an economic perspective, there is still plenty of demand in the market for non-cutting edge node sizes, i.e. in embedded the priority is more around the fact that the chip is not changed from a specific design that was made 4 years ago than having the latest and greatest.

From a strategic perspective, it is absolutely important that you have the capability to build chips in your country, of a reasonably modern node size. You need this for weapons manufacturing, for a working government apparatus (governments use computers now), for sending messages to your population. And if China bombs Taiwan, then this fab will become the cutting edge, instead of you having zero chip manufacturing capabilities.

It's a start. Certainly miles beyond where US domestic fabrication stands today.

Indeed. There are zero domestic use cases for that ancient 4nm tech…

How many consumers or even servers are running latest generation? 4nm indeed seems like it would have legs for a long time.

Exactly. I get how having the latest and greatest would be ideal, but scoffing at 4nm seems silly to me.

Reply on news.ycombinator.com